diff --git a/tests/eval/test_utils.py b/tests/eval/test_utils_eval.py
similarity index 100%
rename from tests/eval/test_utils.py
rename to tests/eval/test_utils_eval.py
diff --git a/tests/train/test_utils.py b/tests/train/test_utils.py
new file mode 100644
index 00000000..2a29a4b9
--- /dev/null
+++ b/tests/train/test_utils.py
@@ -0,0 +1,35 @@
+import torch
+from datasets import Dataset
+
+from delphi.train.utils import gen_minibatches
+
+
+def test_gen_minibatches():
+    DS_SIZE = 6
+    SEQ_LEN = 5
+    NUM_MINIBATCHES = 3
+    MINIBATCH_SIZE = DS_SIZE // NUM_MINIBATCHES
+    FEATURE_NAME = "tokens"
+    dataset = Dataset.from_dict(
+        {
+            FEATURE_NAME: [list(range(i, i + SEQ_LEN)) for i in range(DS_SIZE)],
+        },
+    )
+    dataset.set_format(type="torch")
+    indices = list(range(DS_SIZE - 1, -1, -1))
+    minibatches = gen_minibatches(
+        dataset=dataset,
+        batch_size=DS_SIZE,
+        num_minibatches=NUM_MINIBATCHES,
+        step=0,
+        indices=indices,
+        device=torch.device("cpu"),
+        feature_name=FEATURE_NAME,
+    )
+    minibatches = list(minibatches)
+    assert len(minibatches) == NUM_MINIBATCHES
+    shuffled_ds = dataset[FEATURE_NAME][indices]  # type: ignore
+    for i, minibatch in enumerate(minibatches):
+        assert minibatch.shape == (MINIBATCH_SIZE, SEQ_LEN)
+        expected_mb = shuffled_ds[i * MINIBATCH_SIZE : (i + 1) * MINIBATCH_SIZE]
+        assert torch.all(minibatch == expected_mb)